When neural network forecasts missile spare parts, redundant parameter is prone to making event too long and getting into part optimization, in order to solve these problems, established a consumption forecasting model of missile spare parts based on rough sets and BP neural network. Firstly, consumption information of missile spare parts was abstracted and made into decision-making table; Secondly, simplified original information table and deleted redundant property and property value by rough sets theory; Lastly, the simplified influence factor value was put into BP neural network to carry out training and forecasting. The example results proved the consumption forecasting method reduced greatly convergence time of neural network, improved forecast precision, and afforded a new way for consumption forecasting of missile spare parts.